semantics and RDF(S)

In the process of working on the design of DAML+OIL, I have had extensive
dealings with RDF and RDF Schema.  In this message I present the biggest
problem that I see with RDF and RDF Schema.


The problem that I see has to do with the meaning of constructs in RDF and
RDF Schema.  Although RDF and RDF Schema are supposed to provide machine
understandability for the WWW they have no semantics for most of their
constructs.

Sure there is a mapping from RDF and RDF Schema documents into triples, but
this is nowhere near a semantics for RDF and RDF Schema.  A semantics would
provide a meaning for RDF containers---one that would settle matters of
whether two bags are the same, whether a container can contain itself, what
alternative means, whether a container can have missing elements, and what
distributive statements mean.  A semantics would provide a meaning for
reifications.  It would answer questions about the relationship between a
statement and its reification.  A semantics would provide a notion of
equality for rdfs:Classes.

Because most RDF and RDF Schema constructs do not have a firm meaning, any
use of them is on extremely shaky ground.  How can I build any
semantically-meaningful construct on top of RDF containers when I don't
know that RDF containers mean?  How can I build a notion of cardinality
when there is no notion of equality for much of RDF and RDF Schema?


In the effort of producing DAML+OIL some of us have tried to provide a
meaning for part of RDF and RDF Schema.  I have gone a minimalist route,
and as part of the denotational semantics for DAML+OIL I have created
some semantic mappings for RDF and RDF Schema.  Richard Fikes and Deborah
McGuinness have created an axiomatic semantics for part of RDF and RDF
Schema.  Neither of these semantics is anywhere near complete---they just
provide enough to get (part way towards) a semantics for DAML+OIL.

I strongly believe that RDF and RDF Schema need a full, formal, and
unambiguous semantics.  Without such a semantics people working with RDF
and RDF Schema will be doomed to endless ``How many statings can dance on
the head of a reified statement'' debates.  Although such debates can be
very amusing, they don't produce much in the way of useful knowledge.  Even
worse, programs that deal with RDF and RDF Schema will incorporate
different ideas on the meanings of RDF and RDF Schema constructs, and will
thus not be able to reliably communicate with each other.

Producing a good semantics for the constructs in RDF and RDF Schema will
not be easy.  It may be that parts of RDF and RDF Schema will not fit into
the semantics and may have to be modified or abandoned.  If that is the
case, well I'm willing to excise the diseased part of a formalism to save
it. 


Comments on this are very welcome.  Even more welcome would be efforts to
come up with a semantics for RDF and RDF Schema.  (I don't have one handy,
by the way.  I have ideas for semantics for parts of RDF and RDF Schema,
but certainly not the entirety of them.)

Peter Patel-Schneider
Bell Labs Research
pfps@research.bell-labs.com

Received on Friday, 12 January 2001 08:43:19 UTC